1/*- 2 * Copyright (c) 2003 Peter Wemm. 3 * Copyright (c) 1990 The Regents of the University of California. 4 * All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * William Jolitz. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * $FreeBSD$ 34 */ 35 36#include <machine/asmacros.h> 37#include <machine/specialreg.h> 38 39#include "assym.inc" 40#include "opt_sched.h" 41 42/*****************************************************************************/ 43/* Scheduling */ 44/*****************************************************************************/ 45 46 .text 47 48/* 49 * cpu_throw() 50 * 51 * This is the second half of cpu_switch(). It is used when the current 52 * thread is either a dummy or slated to die, and we no longer care 53 * about its state. This is only a slight optimization and is probably 54 * not worth it anymore. Note that we need to clear the pm_active bits so 55 * we do need the old proc if it still exists. 56 * %rdi = oldtd 57 * %rsi = newtd 58 */ 59ENTRY(cpu_throw) 60 movq %rsi,%r12 61 movq %rsi,%rdi 62 call pmap_activate_sw 63 jmp sw1 64END(cpu_throw) 65 66/* 67 * cpu_switch(old, new, mtx) 68 * 69 * Save the current thread state, then select the next thread to run 70 * and load its state. 71 * %rdi = oldtd 72 * %rsi = newtd 73 * %rdx = mtx 74 */ 75ENTRY(cpu_switch) 76 /* Switch to new thread. First, save context. */ 77 leaq TD_MD_PCB(%rdi),%r8 78 79 movq (%rsp),%rax /* Hardware registers */ 80 movq %r15,PCB_R15(%r8) 81 movq %r14,PCB_R14(%r8) 82 movq %r13,PCB_R13(%r8) 83 movq %r12,PCB_R12(%r8) 84 movq %rbp,PCB_RBP(%r8) 85 movq %rsp,PCB_RSP(%r8) 86 movq %rbx,PCB_RBX(%r8) 87 movq %rax,PCB_RIP(%r8) 88 89 testl $PCB_FULL_IRET,PCB_FLAGS(%r8) 90 jnz 2f 91 orl $PCB_FULL_IRET,PCB_FLAGS(%r8) 92 testl $TDP_KTHREAD,TD_PFLAGS(%rdi) 93 jnz 2f 94 testb $CPUID_STDEXT_FSGSBASE,cpu_stdext_feature(%rip) 95 jz 2f 96 movl %fs,%eax 97 cmpl $KUF32SEL,%eax 98 jne 1f 99 rdfsbase %rax 100 movq %rax,PCB_FSBASE(%r8) 1011: movl %gs,%eax 102 cmpl $KUG32SEL,%eax 103 jne 2f 104 movq %rdx,%r12 105 movl $MSR_KGSBASE,%ecx /* Read user gs base */ 106 rdmsr 107 shlq $32,%rdx 108 orq %rdx,%rax 109 movq %rax,PCB_GSBASE(%r8) 110 movq %r12,%rdx 111 1122: 113 testl $PCB_DBREGS,PCB_FLAGS(%r8) 114 jnz store_dr /* static predict not taken */ 115done_store_dr: 116 117 /* have we used fp, and need a save? */ 118 cmpq %rdi,PCPU(FPCURTHREAD) 119 jne 2f 120 movq PCB_SAVEFPU(%r8),%r8 121 clts 122 cmpl $0,use_xsave(%rip) 123 jne 1f 124 fxsave (%r8) 125 jmp 2f 1261: movq %rdx,%rcx 127 movl xsave_mask,%eax 128 movl xsave_mask+4,%edx 129 .globl ctx_switch_xsave 130ctx_switch_xsave: 131 /* This is patched to xsaveopt if supported, see fpuinit_bsp1() */ 132 xsave (%r8) 133 movq %rcx,%rdx 1342: 135 /* Save is done. Now fire up new thread. Leave old vmspace. */ 136 movq %rsi,%r12 137 movq %rdi,%r13 138 movq %rdx,%r15 139 movq %rsi,%rdi 140 callq pmap_activate_sw 141 movq %r15,TD_LOCK(%r13) /* Release the old thread */ 142sw1: 143 leaq TD_MD_PCB(%r12),%r8 144#if defined(SCHED_ULE) && defined(SMP) 145 movq $blocked_lock, %rdx 146 movq TD_LOCK(%r12),%rcx 147 cmpq %rcx, %rdx 148 je sw1wait 149sw1cont: 150#endif 151 /* 152 * At this point, we've switched address spaces and are ready 153 * to load up the rest of the next context. 154 */ 155 156 /* Skip loading LDT and user fsbase/gsbase for kthreads */ 157 testl $TDP_KTHREAD,TD_PFLAGS(%r12) 158 jnz do_kthread 159 160 /* 161 * Load ldt register 162 */ 163 movq TD_PROC(%r12),%rcx 164 cmpq $0, P_MD+MD_LDT(%rcx) 165 jne do_ldt 166 xorl %eax,%eax 167ld_ldt: lldt %ax 168 169 /* Restore fs base in GDT */ 170 movl PCB_FSBASE(%r8),%eax 171 movq PCPU(FS32P),%rdx 172 movw %ax,2(%rdx) 173 shrl $16,%eax 174 movb %al,4(%rdx) 175 shrl $8,%eax 176 movb %al,7(%rdx) 177 178 /* Restore gs base in GDT */ 179 movl PCB_GSBASE(%r8),%eax 180 movq PCPU(GS32P),%rdx 181 movw %ax,2(%rdx) 182 shrl $16,%eax 183 movb %al,4(%rdx) 184 shrl $8,%eax 185 movb %al,7(%rdx) 186 187do_kthread: 188 /* Do we need to reload tss ? */ 189 movq PCPU(TSSP),%rax 190 movq PCB_TSSP(%r8),%rdx 191 movq PCPU(PRVSPACE),%r13 192 addq $PC_COMMONTSS,%r13 193 testq %rdx,%rdx 194 cmovzq %r13,%rdx 195 cmpq %rax,%rdx 196 jne do_tss 197done_tss: 198 movq TD_MD_STACK_BASE(%r12),%r9 199 movq %r9,PCPU(RSP0) 200 movq %r8,PCPU(CURPCB) 201 movq PCPU(PTI_RSP0),%rax 202 cmpq $~0,PCPU(UCR3) 203 cmove %r9,%rax 204 movq %rax,TSS_RSP0(%rdx) 205 movq %r12,PCPU(CURTHREAD) /* into next thread */ 206 207 /* Test if debug registers should be restored. */ 208 testl $PCB_DBREGS,PCB_FLAGS(%r8) 209 jnz load_dr /* static predict not taken */ 210done_load_dr: 211 212 /* Restore context. */ 213 movq PCB_R15(%r8),%r15 214 movq PCB_R14(%r8),%r14 215 movq PCB_R13(%r8),%r13 216 movq PCB_R12(%r8),%r12 217 movq PCB_RBP(%r8),%rbp 218 movq PCB_RSP(%r8),%rsp 219 movq PCB_RBX(%r8),%rbx 220 movq PCB_RIP(%r8),%rax 221 movq %rax,(%rsp) 222 movq PCPU(CURTHREAD),%rdi 223 call fpu_activate_sw 224 cmpb $0,cpu_flush_rsb_ctxsw(%rip) 225 jne rsb_flush 226 ret 227 228 /* 229 * We order these strangely for several reasons. 230 * 1: I wanted to use static branch prediction hints 231 * 2: Most athlon64/opteron cpus don't have them. They define 232 * a forward branch as 'predict not taken'. Intel cores have 233 * the 'rep' prefix to invert this. 234 * So, to make it work on both forms of cpu we do the detour. 235 * We use jumps rather than call in order to avoid the stack. 236 */ 237 238store_dr: 239 movq %dr7,%rax /* yes, do the save */ 240 movq %dr0,%r15 241 movq %dr1,%r14 242 movq %dr2,%r13 243 movq %dr3,%r12 244 movq %dr6,%r11 245 movq %r15,PCB_DR0(%r8) 246 movq %r14,PCB_DR1(%r8) 247 movq %r13,PCB_DR2(%r8) 248 movq %r12,PCB_DR3(%r8) 249 movq %r11,PCB_DR6(%r8) 250 movq %rax,PCB_DR7(%r8) 251 andq $0x0000fc00, %rax /* disable all watchpoints */ 252 movq %rax,%dr7 253 jmp done_store_dr 254 255load_dr: 256 movq %dr7,%rax 257 movq PCB_DR0(%r8),%r15 258 movq PCB_DR1(%r8),%r14 259 movq PCB_DR2(%r8),%r13 260 movq PCB_DR3(%r8),%r12 261 movq PCB_DR6(%r8),%r11 262 movq PCB_DR7(%r8),%rcx 263 movq %r15,%dr0 264 movq %r14,%dr1 265 /* Preserve reserved bits in %dr7 */ 266 andq $0x0000fc00,%rax 267 andq $~0x0000fc00,%rcx 268 movq %r13,%dr2 269 movq %r12,%dr3 270 orq %rcx,%rax 271 movq %r11,%dr6 272 movq %rax,%dr7 273 jmp done_load_dr 274 275do_tss: movq %rdx,PCPU(TSSP) 276 movq %rdx,%rcx 277 movq PCPU(TSS),%rax 278 movw %cx,2(%rax) 279 shrq $16,%rcx 280 movb %cl,4(%rax) 281 shrq $8,%rcx 282 movb %cl,7(%rax) 283 shrq $8,%rcx 284 movl %ecx,8(%rax) 285 movb $0x89,5(%rax) /* unset busy */ 286 movl $TSSSEL,%eax 287 ltr %ax 288 jmp done_tss 289 290do_ldt: movq PCPU(LDT),%rax 291 movq P_MD+MD_LDT_SD(%rcx),%rdx 292 movq %rdx,(%rax) 293 movq P_MD+MD_LDT_SD+8(%rcx),%rdx 294 movq %rdx,8(%rax) 295 movl $LDTSEL,%eax 296 jmp ld_ldt 297END(cpu_switch) 298 299/* 300 * savectx(pcb) 301 * Update pcb, saving current processor state. 302 */ 303ENTRY(savectx) 304 /* Save caller's return address. */ 305 movq (%rsp),%rax 306 movq %rax,PCB_RIP(%rdi) 307 308 movq %rbx,PCB_RBX(%rdi) 309 movq %rsp,PCB_RSP(%rdi) 310 movq %rbp,PCB_RBP(%rdi) 311 movq %r12,PCB_R12(%rdi) 312 movq %r13,PCB_R13(%rdi) 313 movq %r14,PCB_R14(%rdi) 314 movq %r15,PCB_R15(%rdi) 315 316 movq %cr0,%rax 317 movq %rax,PCB_CR0(%rdi) 318 movq %cr2,%rax 319 movq %rax,PCB_CR2(%rdi) 320 movq %cr3,%rax 321 movq %rax,PCB_CR3(%rdi) 322 movq %cr4,%rax 323 movq %rax,PCB_CR4(%rdi) 324 325 movq %dr0,%rax 326 movq %rax,PCB_DR0(%rdi) 327 movq %dr1,%rax 328 movq %rax,PCB_DR1(%rdi) 329 movq %dr2,%rax 330 movq %rax,PCB_DR2(%rdi) 331 movq %dr3,%rax 332 movq %rax,PCB_DR3(%rdi) 333 movq %dr6,%rax 334 movq %rax,PCB_DR6(%rdi) 335 movq %dr7,%rax 336 movq %rax,PCB_DR7(%rdi) 337 338 movl $MSR_FSBASE,%ecx 339 rdmsr 340 movl %eax,PCB_FSBASE(%rdi) 341 movl %edx,PCB_FSBASE+4(%rdi) 342 movl $MSR_GSBASE,%ecx 343 rdmsr 344 movl %eax,PCB_GSBASE(%rdi) 345 movl %edx,PCB_GSBASE+4(%rdi) 346 movl $MSR_KGSBASE,%ecx 347 rdmsr 348 movl %eax,PCB_KGSBASE(%rdi) 349 movl %edx,PCB_KGSBASE+4(%rdi) 350 movl $MSR_EFER,%ecx 351 rdmsr 352 movl %eax,PCB_EFER(%rdi) 353 movl %edx,PCB_EFER+4(%rdi) 354 movl $MSR_STAR,%ecx 355 rdmsr 356 movl %eax,PCB_STAR(%rdi) 357 movl %edx,PCB_STAR+4(%rdi) 358 movl $MSR_LSTAR,%ecx 359 rdmsr 360 movl %eax,PCB_LSTAR(%rdi) 361 movl %edx,PCB_LSTAR+4(%rdi) 362 movl $MSR_CSTAR,%ecx 363 rdmsr 364 movl %eax,PCB_CSTAR(%rdi) 365 movl %edx,PCB_CSTAR+4(%rdi) 366 movl $MSR_SF_MASK,%ecx 367 rdmsr 368 movl %eax,PCB_SFMASK(%rdi) 369 movl %edx,PCB_SFMASK+4(%rdi) 370 371 sgdt PCB_GDT(%rdi) 372 sidt PCB_IDT(%rdi) 373 sldt PCB_LDT(%rdi) 374 str PCB_TR(%rdi) 375 376 movl $1,%eax 377 ret 378END(savectx) 379 380/* 381 * resumectx(pcb) 382 * Resuming processor state from pcb. 383 */ 384ENTRY(resumectx) 385 /* Switch to KPML4phys. */ 386 movq KPML4phys,%rax 387 movq %rax,%cr3 388 389 /* Force kernel segment registers. */ 390 movl $KDSEL,%eax 391 movw %ax,%ds 392 movw %ax,%es 393 movw %ax,%ss 394 movl $KUF32SEL,%eax 395 movw %ax,%fs 396 movl $KUG32SEL,%eax 397 movw %ax,%gs 398 399 movl $MSR_FSBASE,%ecx 400 movl PCB_FSBASE(%rdi),%eax 401 movl 4 + PCB_FSBASE(%rdi),%edx 402 wrmsr 403 movl $MSR_GSBASE,%ecx 404 movl PCB_GSBASE(%rdi),%eax 405 movl 4 + PCB_GSBASE(%rdi),%edx 406 wrmsr 407 movl $MSR_KGSBASE,%ecx 408 movl PCB_KGSBASE(%rdi),%eax 409 movl 4 + PCB_KGSBASE(%rdi),%edx 410 wrmsr 411 412 /* Restore EFER one more time. */ 413 movl $MSR_EFER,%ecx 414 movl PCB_EFER(%rdi),%eax 415 wrmsr 416 417 /* Restore fast syscall stuff. */ 418 movl $MSR_STAR,%ecx 419 movl PCB_STAR(%rdi),%eax 420 movl 4 + PCB_STAR(%rdi),%edx 421 wrmsr 422 movl $MSR_LSTAR,%ecx 423 movl PCB_LSTAR(%rdi),%eax 424 movl 4 + PCB_LSTAR(%rdi),%edx 425 wrmsr 426 movl $MSR_CSTAR,%ecx 427 movl PCB_CSTAR(%rdi),%eax 428 movl 4 + PCB_CSTAR(%rdi),%edx 429 wrmsr 430 movl $MSR_SF_MASK,%ecx 431 movl PCB_SFMASK(%rdi),%eax 432 wrmsr 433 434 /* Restore CR0, CR2, CR4 and CR3. */ 435 movq PCB_CR0(%rdi),%rax 436 movq %rax,%cr0 437 movq PCB_CR2(%rdi),%rax 438 movq %rax,%cr2 439 movq PCB_CR4(%rdi),%rax 440 movq %rax,%cr4 441 movq PCB_CR3(%rdi),%rax 442 movq %rax,%cr3 443 444 /* Restore descriptor tables. */ 445 lidt PCB_IDT(%rdi) 446 lldt PCB_LDT(%rdi) 447 448#define SDT_SYSTSS 9 449#define SDT_SYSBSY 11 450 451 /* Clear "task busy" bit and reload TR. */ 452 movq PCPU(TSS),%rax 453 andb $(~SDT_SYSBSY | SDT_SYSTSS),5(%rax) 454 movw PCB_TR(%rdi),%ax 455 ltr %ax 456 457#undef SDT_SYSTSS 458#undef SDT_SYSBSY 459 460 /* Restore debug registers. */ 461 movq PCB_DR0(%rdi),%rax 462 movq %rax,%dr0 463 movq PCB_DR1(%rdi),%rax 464 movq %rax,%dr1 465 movq PCB_DR2(%rdi),%rax 466 movq %rax,%dr2 467 movq PCB_DR3(%rdi),%rax 468 movq %rax,%dr3 469 movq PCB_DR6(%rdi),%rax 470 movq %rax,%dr6 471 movq PCB_DR7(%rdi),%rax 472 movq %rax,%dr7 473 474 /* Restore other callee saved registers. */ 475 movq PCB_R15(%rdi),%r15 476 movq PCB_R14(%rdi),%r14 477 movq PCB_R13(%rdi),%r13 478 movq PCB_R12(%rdi),%r12 479 movq PCB_RBP(%rdi),%rbp 480 movq PCB_RSP(%rdi),%rsp 481 movq PCB_RBX(%rdi),%rbx 482 483 /* Restore return address. */ 484 movq PCB_RIP(%rdi),%rax 485 movq %rax,(%rsp) 486 487 xorl %eax,%eax 488 ret 489END(resumectx) 490 491/* Wait for the new thread to become unblocked */ 492#if defined(SCHED_ULE) && defined(SMP) 493sw1wait: 4941: 495 pause 496 movq TD_LOCK(%r12),%rcx 497 cmpq %rcx, %rdx 498 je 1b 499 jmp sw1cont 500#endif 501